Chapter 6. Order Statistics and Quantiles. 6.1 Extreme Order Statistics

Transcription

1 Chapter 6 Order Statistics and Quantiles 61 Extreme Order Statistics Suppose we have a finite sample X 1,, X n Conditional on this sample, we define the values X 1),, X n) to be a permutation of X 1,, X n such that X 1) X 2) X n) We call X i) the ith order statistic of the sample In much of this chapter, we will be concerned with the marginal distributions of these order statistics Even though the notation X i) does not explicitly use the sample size n, the distribution of X i) depends essentially on n For this reason, some textbooks use slightly more complicated notation such as X 1:n), X 2;n),, X n;n) for the order statistics of a sample We choose to use the simpler notation here, though we will always understand the sample size to be n The asymptotic distributions of order statistics at the extremes of a sample may be derived without any specialized knowledge other than the limit formula 1 + c ) n e c as n 61) n and its generalization 1 + c ) bn n e c if c n c and b n 62) b n [see Exercise 17b)] Recall that by asymptotic distribution of X 1), we mean sequences k n and a n, along with a nondegenerate random variable X, such that k n X 1) a n ) d X This section consists mostly of a series of illustrative examples 96

2 Example 61 Suppose X 1,, X n are independent and identically distributed uniform0,1) random variables What is the asymptotic distribution of X n)? Since X n) t if and only if X 1 t, X 2 t,, and X n t, by independence we have { 0 if t 0 P X n) t) = t n if 0 < t < 1 63) 1 if t 1 From Equation 63), it is apparent that X n) P 1, though this limit statement does not fully reveal the asymptotic distribution of X n) We desire sequences k n and a n such that k n X n) a n ) has a nondegenerate limiting distribution Evidently, we should expect a n = 1, a fact we shall rederive below Computing the distribution function of k n X n) a n ) directly, we find } F u) = P {k n X n) a n ) u} = P {X n) ukn + a n as long as k n > 0 Therefore, we see that ) n u F u) = + a n for 0 < u + a n < 1 64) k n k n We would like this expression to tend to a limit involving only u as n Keeping expression 62 in mind, we take a n = 1 and k n = n so that F u) = 1 + u/n) n, which tends to e u However, we are not quite finished, since we have not determined which values of u make the above limit valid Equation 64 required that 0 < a n + u/k n ) < 1, which in this case becomes 1 < u/n < 0 This means u may be any negative real number, since for any u < 0, 1 < u/n < 0 for all n > u We conclude that if the random variable U has distribution function { expu) if u 0 F u) = 1 if u > 0, then nx n) 1) d U Since U is simply a standard exponential random variable, we may also write n1 X n) ) d Exponential1) Example 62 Suppose X 1, X 2, are independent and identically distributed exponential random variables with mean 1 What is the asymptotic distribution of X n)? 97

3 As in Equation 63, if u/k n + a n > 0 then P { k n X n) a n ) u } ) = P X n) ukn + a n = { 1 exp a n u )} n k n Taking a n = log n and k n = 1, the rightmost expression above simplifies to } n {1 e u, n which has limit exp e u ) The condition u/k n + a n > 0 becomes u + log n > 0, which is true for all u R as long as n > exp u) Therefore, we conclude that X n) log n d U, where P U u) def = exp{ exp u)} for all u 65) The distribution of U in Equation 65 is known as the extreme value distribution or the Gumbel distribution In Examples 61 and 62, we derived the asymptotic distribution of a maximum from a simple random sample We did this using only the definition of convergence in distribution without relying on any results other than expression 62 In a similar way, we may derive the joint asymptotic distribution of multiple order statistics, as in the following example Example 63 Range of uniform sample: Let X 1,, X n be a simple random sample from Uniform0, 1) Let R n = X n) X 1) denote the range of the sample What is the asymptotic distribution of R n? To answer this question, we begin by finding the joint asymptotic distribution of X n), X 1) ), as follows For sequences k n and l n, as yet unspecified, consider P k n X 1) > x and l n 1 X n) ) > y) = P X 1) > x/k n and X n) < 1 y/l n ) = P x/k n < X 1) < < X n) < 1 y/l n ), where we have assumed that k n and l n are positive Since the probability above is simply the probability that the entire sample is to be found in the interval x/k n, 1 y/l n ), we conclude that as long as 0 < x k n < 1 y l n < 1, 66) we have P k n X 1) > x and l n 1 X n) ) > y) = 1 y l n x k n ) n 98

4 Expression 61) suggests that we set k n = l n = n, resulting in P nx 1) > x and n1 X n) ) > y) = 1 y n n) x n Expression 66 becomes 0 < x n < 1 y n < 1, which is satisfied for large enough n if and only if x and y are both positive We conclude that for x > 0, y > 0, P nx 1) > x and n1 X n) ) > y) e x e y Since this is the joint distribution of independent standard exponential random variables, say, Y 1 and Y 2, we conclude that ) ) nx1) d Y1 n1 X n) ) Therefore, applying the continuous function fa, b) = a + b to both sides gives n1 X n) + X 1) ) = n1 R n ) d Y 1 + Y 2 Gamma2, 1) Let us consider a different example in which the asymptotic joint distribution does not involve independent random variables Example 64 As in Example 63, let X 1,, X n be independent and identically distributed from uniform0, 1) What is the joint asymptotic distribution of X n 1) and X n)? Proceeding as in Example 63, we obtain [ ) )] n1 Xn 1) ) x P > = P n1 X n) ) y Y 2 X n 1) < 1 x n and X n) < 1 y n ) 67) We consider two separate cases: If 0 < x < y, then the right hand side of 67) is simply P X n) < 1 y/n), which converges to e y as in Example 61 On the other hand, if 0 < y < x, then P X n 1) < 1 x n and X n) < 1 y ) = P X n) < 1 x ) n n 99 = +P 1 x n e x 1 + x y) X n 1) < 1 x n < X n) < 1 y n ) n + n 1 x ) n 1 x n n n) y )

5 The second equality above arises because X n 1) < a < X n) < b if and only if exactly n 1 of the X i are less than a and exactly one is between a and b We now know the joint asymptotic distribution of n1 X n 1) ) and n1 X n 1) ); but can we describe this joint distribution in a simple way? Suppose that Y 1 and Y 2 are independent standard exponential variables Consider the joint distribution of Y 1 and Y 1 + Y 2 : If 0 < x < y, then P Y 1 + Y 2 > x and Y 1 > y) = P Y 1 > y) = e y On the other hand, if 0 < y < x, then P Y 1 + Y 2 > x and Y 1 > y) = P Y 1 > max{y, x Y 2 }) = E e max{y,x Y 2} Therefore, we conclude that ) n1 Xn 1) ) n1 X n) ) = e y P y > x Y 2 ) + = e x 1 + x y) d Y1 + Y 2 Y 1 ) x y 0 e t x e t dt Notice that marginally, we have shown that n1 X n 1) ) d Gamma2, 1) Recall that if F is a continuous, invertible distribution function and U is a standard uniform random variable, then F 1 U) F The proof is immediate, since P {F 1 U) t} = P {U F t)} = F t) We may use this fact in conjunction with the result of Example 64 as in the following example Example 65 Suppose X 1,, X n are independent standard exponential random variables What is the joint asymptotic distribution of X n 1), X n) )? The distribution function of a standard exponential distribution is F t) = 1 e t, whose inverse is F 1 u) = log1 u) Therefore, ) ) log1 Un 1) ) d Xn 1) =, log1 U n) ) X n) where = d means has the same distribution Thus, [ log n1 Un 1) ) ] ) log [ n1 U n) ) ] d = Xn 1) log n X n) log n We conclude by the result of Example 64 that ) ) Xn 1) log n d logy1 + Y 2 ), X n) log n log Y 1 where Y 1 and Y 2 are independent standard exponential variables 100 )

6 Exercises for Section 61 Exercise 61 For a given n, let X 1,, X n be independent and identically distributed with distribution function P X i t) = t3 + θ 3 2θ 3 for t [ θ, θ] Let X 1) denote the first order statistic from the sample of size n; that is, X 1) is the smallest of the X i a) Prove that X 1) is consistent for θ b) Prove that nθ + X 1) ) d Y, where Y is a random variable with an exponential distribution Find E Y ) in terms of θ c) For a fixed α, define δ α,n = 1 + α ) X 1) n Find, with proof, α such that n θ + δ α,n) d Y E Y ), where Y is the same random variable as in part b) d) Compare the two consistent θ-estimators δ α,n and X 1) empirically as follows For n {10 2, 10 3, 10 4 }, take θ = 1 and simulate 1000 samples of size n from the distribution of X i From these 1000 samples, estimate the bias and mean squared error of each estimator Which of the two appears better? Do your empirical results agree with the theoretical results in parts c) and d)? Exercise 62 Let X 1, X 2, be independent uniform 0, θ) random variables Let X n) = max{x 1,, X n } and consider the three estimators δn 0 = X n) δn 1 = n ) 2 n n 1 X n) δn 2 = X n) n 1 a) Prove that each estimator is consistent for θ 101

7 b) Perform an empirical comparison of these three estimators for n = 10 2, 10 3, 10 4 Use θ = 1 and simulate 1000 samples of size n from uniform 0, 1) From these 1000 samples, estimate the bias and mean squared error of each estimator Which one of the three appears to be best? c) Find the asymptotic distribution of nθ δ i n) for i = 0, 1, 2 Based on your results, which of the three appears to be the best estimator and why? For the latter question, don t attempt to make a rigorous mathematical argument; simply give an educated guess) Exercise 63 Find, with proof, the asymptotic distribution of X n) if X 1,, X n are independent and identically distributed with each of the following distributions That is, find k n, a n, and a nondegenerate random variable X such that k n X n) a n ) d X) a) Beta3, 1) with distribution function F x) = x 2 for x 0, 1) b) Standard logistic with distribution function F x) = e x /1 + e x ) Exercise 64 Let X 1,, X n be independent uniform0, 1) random variables Find the joint asymptotic distribution of [ nx 2), n1 X n 1) ) ] Hint: To find a probability such as P a < X 2) < X n) < b), consider the trinomial distribution with parameters [n; a, b a, 1 b)] and note that the probability in question is the same as the probability that the numbers in the first and third categories are each 1 Exercise 65 Let X 1,, X n be a simple random sample from the distribution function F x) = [1 1/x)]I{x > 1} a) Find the joint asymptotic distribution of X n 1) /n, X n) /n) Hint: Proceed as in Example 65 b) Find the asymptotic distribution of X n 1) /X n) Exercise 66 If X 1,, X n are independent and identically distributed uniform0, 1) variables, prove that X 1) /X 2) d uniform0, 1) Exercise 67 Let X 1,, X n be independent uniform0, 2θ) random variables a) Let M = X 1) + X n) )/2 Find the asymptotic distribution of nm θ) b) Compare the asymptotic performance of the three estimators M, X n, and the sample median X n by considering their relative efficiencies 102

8 c) For n {101, 1001, 10001}, generate 500 samples of size n, taking θ = 1 Keep track of M, X n, and X n for each sample Construct a 3 3 table in which you report the sample variance of each estimator for each value of n Do your simulation results agree with your theoretical results in part b)? Exercise 68 Let X 1,, X n be a simple random sample from a logistic distribution with distribution function F t) = e t/θ /1 + e t/θ ) for all t a) Find the asymptotic distribution of X n) X n 1) Hint: zero Use the fact that log U n) and log U n 1) both converge in probability to b) Based on part a), construct an approximate 95% confidence interval for θ Use the fact that the 025 and 975 quantiles of the standard exponential distribution are and 36889, respectively c) Simulate 1000 samples of size n = 40 with θ = 2 How many confidence intervals contain θ? 62 Sample Quantiles To derive the distribution of sample quantiles, we begin by obtaining the exact distribution of the order statistics of a random sample from a uniform distribution To facilitate this derivation, we begin with a quick review of changing variables Suppose X has density f X x) and Y = gx), where g : R k R k is differentiable and has a well-defined inverse, which we denote by h : R k R k In particular, we have X = h[y]) The density for Y is f Y y) = Det[ hy)] f X [hy)], 68) where Det[ hy)] is the absolute value of the determinant of the k k matrix hy) 621 Uniform Order Statistics We now show that the order statistics of a uniform distribution may be obtained using ratios of gamma random variables Suppose X 1,, X n+1 are independent standard exponential, or Gamma1, 1), random variables For j = 1,, n, define Y j = j i=1 X i n+1 i=1 X 69) i 103

9 We will show that the joint distribution of Y 1,, Y n ) is the same as the joint distribution of the order statistics U 1),, U n) ) of a simple random sample from uniform0, 1) by demonstrating that their joint density function is the same as that of the uniform order statistics, namely n!i{0 < u 1) < < u n) < 1} We derive the joint density of Y 1,, Y n ) as follows As an intermediate step, define Z j = j i=1 X i for j = 1,, n + 1 Then X i = { Zi if i = 1 Z i Z i 1 if i > 1, which means that the gradient of the transformation from Z to X is upper triangular with ones on the diagonal, a matrix whose determinant is one This implies that the density for Z is f Z z) = exp{ z n+1 }I{0 < z 1 < z 2 < < z n+1 } Next, if we define Y n+1 = Z n+1, then we may express Z in terms of Y as Z i = { Yn+1 Y i if i < n + 1 Y n+1 if i = n ) The gradient of the transformation in Equation 610) is lower triangular, with y n+1 along the diagonal except for a 1 in the lower right corner The determinant of this matrix is y n n+1, so the density of Y is f Y y) = y n n+1 exp{ y n+1 }I{y n+1 > 0}I{0 < y 1 < < y n < 1} 611) Equation 611) reveals several things: First, Y 1,, Y n ) is independent of Y n+1 and the marginal distribution of Y n+1 is Gamman + 1, 1) More important for our purposes, the marginal joint density of Y 1,, Y n ) is proportional to I{0 < y 1 < < y n < 1}, which is exactly what we needed to prove We conclude that the vector Y defined in Equation 69) has the same distribution as the vector of order statistics of a simple random sample from uniform0, 1) 622 Uniform Sample Quantiles Using the result of Section 621, we may derive the joint asymptotic distribution of a set of sample quantiles for a uniform simple random sample Suppose we are interested in the p 1 and p 2 quantiles, where 0 < p 1 < p 2 < 1 The following argument may be generalized to obtain the joint asymptotic distribution of any finite number 104

10 of quantiles If U 1,, U n are independent uniform0, 1) random variables, then the p 1 and p 2 sample quantiles may be taken to be the a n th and b n th order statistics, respectively, where def def = 5 + np 1 and b n = 5 + np x is simply x rounded to the nearest integer) a n Next, let A n def = 1 n a n i=1 X i, B n def = 1 n b n i=a n+1 X i, and C n def = 1 n n+1 i=b n+1 X i We proved in Section 621 that U an), U bn)) has the same distribution as ) ga n, B n, C n ) def A n A n + B n =, 612) A n + B n + C n A n + B n + C n The asymptotic distribution of ga n, B n, C n ) may be determined using the delta method if we can determine the joint asymptotic distribution of A n, B n, C n ) A bit of algebra reveals that an nan n An p 1 ) = an np ) 1 n a n a n = ) an nan an 1 + a n np 1 n a n n By the central limit theorem, a n na n /a n 1) d N0, 1) because the X i have mean 1 and variance 1 Furthermore, a n /n p 1 and the rightmost term above goes to 0, so Slutsky s theorem gives n An p 1 ) d N0, p 1 ) Similar arguments apply to B n and to C n Because A n and B n and C n are independent of one another, we may stack them as in Exercise 219 to obtain A n p 1 d n B n p 2 p 1 N 3 0 p , 0 p 2 p 1 0 C n 1 p p 2 by Slutsky s theorem For g : R 3 R 2 defined in Equation 612), we obtain 1 ga, b, c) = b + c c a c a + b + c) 2 a a b Therefore, [ gp 1, p 2 p 1, 1 p 2 )] T 1 p1 p = 1 p 1 1 p 2 1 p 2 p 2 ), 105

11 so the delta method gives { ) Uan) n U bn) p1 p 2 )} { ) d 0 p1 1 p N 2, 1 ) p 1 1 p 2 ) 0 p 1 1 p 2 ) p 2 1 p 2 ) )} 613) The method used above to derive the joint distribution 613) of two sample quantiles may be extended to any number of quantiles; doing so yields the following theorem: Theorem 66 Suppose that for given constants p 1,, p k with 0 < p 1 < < p k < 1, there exist sequences {a 1n },, {a kn } such that for all 1 i k, a in np i n 0 Then if U 1,, U n is a sample from Uniform0,1), n U a1n ) U akn ) p 1 p 1 1 p 1 ) p 1 1 p k ) d N k 0, p k 0 p 1 1 p k ) p k 1 p k ) Note that for i < j, both the i, j) and j, i) entries in the covariance matrix above equal p i 1 p j ) and p j 1 p i ) never occurs in the matrix 623 General sample quantiles Let F x) be the distribution function for a random variable X The quantile function F u) of Definition 313 is nondecreasing on 0, 1) and it has the property that F U) has the same distribution as X for U uniform0, 1) This property follows from Lemma 314, since P [F U) x] = P [U F x)] = F x) for all x Since F u) is nondecreasing, it preserves ordering; thus, if X 1,, X n is a random sample from F x), then X 1),, X n) ) d = [ F U 1) ),, F U n) ) ] The symbol d = means has the same distribution as ) Now, suppose that at some point ξ, the derivative F ξ) exists and is positive Then F x) must be continuous and strictly increasing in a neighborhood of ξ This implies that in this 106

12 neighborhood, F x) has a well-defined inverse, which must be differentiable at the point p def = F ξ) If F 1 u) denotes the inverse that exists in a neighborhood of p, then df 1 p) dp = 1 F ξ) 614) Equation 614) may be derived by differentiating the equation F [F 1 p)] = p Note also that whenever the inverse F 1 u) exists, it must coincide with the quantile function F u) Thus, the condition that F ξ) exists and is positive is a sufficient condition to imply that F u) is differentiable at p This differentiability is important: If we wish to transform the uniform order statistics U ain ) of Theorem 66 into order statistics X ain ) using the quantile function F u), the delta method requires the differentiability of F u) at each of the points p 1,, p k The delta method, along with Equation 614), yields the following corollary of Theorem 66: Theorem 67 Let X 1,, X n be a simple random sample from a distribution function F x) such that F x) is differentiable at each of the points ξ 1 < < ξ k and F ξ i ) > 0 for all i Denote F ξ i ) by p i Then under the assumptions of Theorem 66, p X 1 1 p 1 ) p a1n ) ξ 1 F d n N k 0 ξ p k ) ) 2 F ξ 1 )F ξ k ), X akn ) ξ k 0 p 1 1 p k ) p F ξ 1 )F ξ k k 1 p k ) ) F ξ k ) 2 Exercises for Section 62 Exercise 69 Let X 1 be Uniform0, 2π) and let X 2 be standard exponential, independent of X 1 Find the joint distribution of Y 1, Y 2 ) = 2X 2 cos X 1, 2X 2 sin X 1 ) Note: Since log U has a standard exponential distribution if U uniform0, 1), this problem may be used to simulate normal random variables using simulated uniform random variables Exercise 610 Suppose X 1,, X n is a simple random sample with P X i x) = F x θ), where F x) is symmetric about zero We wish to estimate θ by Q p + Q 1 p )/2, where Q p and Q 1 p are the p and 1 p sample quantiles, respectively Find the smallest possible asymptotic variance for the estimator and the p for which it is achieved for each of the following forms of F x): a) Standard Cauchy 107

13 b) Standard normal c) Standard double exponential Hint: For at least one of the three parts of this question, you will have to solve for a minimizer numerically Exercise 611 When we use a boxplot to assess the symmetry of a distribution, one of the main things we do is visually compare the lengths of Q 3 Q 2 and Q 2 Q 1, where Q i denotes the ith sample quartile a) Given a random sample of size n from N0, 1), find the asymptotic distribution of Q 3 Q 2 ) Q 2 Q 1 ) b) Repeat part a) if the sample comes from a standard logistic distribution c) Using 1000 simulations from each distribution, use graphs to assess the accuracy of each of the asymptotic approximations above for n = 5 and n = 13 For a sample of size 4n + 1, define Q i to be the in + 1)th order statistic) For each value of n and each distribution, plot the empirical distribution function against the theoretical limiting distribution function Exercise 612 Let X 1,, X n be a random sample from Uniform0, 2θ) Find the asymptotic distributions of the median, the midquartile range, and 2X 3 3n/4 The midquartile range is the mean of the 1st and 3rd quartiles) Compare these three estimates of θ based on their variances 108